Shorter Recessions*

Felipe Larraín B. and Se Kyu Choi**

Pontificia Universidad Católica de Chile and University of Pennsylvania

August 2009

Correspondence address: Casilla 76, Correo 17, Santiago, Chile

Telephone number: 56-2-3544318
Fax number: 56-2-5532377

E-mail: flarrainb@puc.cl, sechoi@econ.upenn.edu

*
Paper presented at the Jornadas Monetarias y Bancarias 2009, Banco Central de la República Argentina.
**
We thank useful comments by Edward Prescott, Juan Bravo, José de Gregorio, Pablo Mendieta, Joaquin
Poblete, José Tavares, Rodrigo Troncoso, and participants at the Latin American Econometric Society
Meeting, and seminars at The American University of Paris and Universidade Nova de Lisboa. The usual
disclaimer applies.
 Shorter Recessions

Abstract
Empirical studies of economic growth across countries are abundant in conclusions, some of them
widely accepted. This is not the case with the empirics of business cycles. Particularly, there exists little
evidence explaining why some countries take more time than others in recovering from recessions.
This paper focuses on recessions. We are not interested in the causes of recessions, but in the
determinants of their length; thus, we study which economic variables accelerate/retard economic recovery.
The estimation of count-data models (Poisson and Negative Binomial) and seemingly unrelated
regressions (SUR), provide clear evidence that more open economies, with diversified exports, deep financial
markets, more flexible labor markets, and an overall efficient property rights system experience shorter
recessions.

Keywords: Recessions, recoveries, business cycles, panel data, count-data models, seemingly unrelated
regressions,.
JEL classification: C25, C33, E30, E32, F02.

2
1. Introduction
Growth is one of the most studied and discussed topics in economics. A vast literature on the subject
avails our knowledge and provides widely accepted conclusions. Since the first studies by Barro and Sala-i-
Martin (1992), the empirical determinants of economic growth have been analyzed rigorously1, giving birth to
a rich literature filled with explanations and policy options to foster it2. In recent years, estimations based on
cross-section data have been substituted with newer and more sophisticated techniques, such as GMM for
dynamic panel data and time-series econometrics, eliminating spurious correlations and providing more
efficient estimates.

Business cycles, in contrast, have been studied in much less detail. In particular, there seems to be
virtually no empirical analyses based on international data comparisons. The lack of a standard procedure to
follow -i.e. linear regression or another estimation technique- along the difficulties arising from the
identification and definition of recessions may provide an answer on why they have not been studied using an
empirical econometric approach. The subject, of course, has not been forgotten by economics. The seminal
work by Kydland and Prescott (1982) and the breakthrough of new computational techniques other than
econometrics, allowed serious study of business cycles based on micro-founded models and mathematical
simulations. Nevertheless, a problem with real business cycle models is the increasing difficulty of modelling
additional variables and effects.

This paper tries to fill a bit of this gap and studies the differences across countries in recovery times
from a recession. This is, to our knowledge, a first formal empirical approach to the subject that incorporates
information of a heterogeneous set of countries. Our analysis does not look for a reduced model containing
the elements that explain and prevent recessions. Its scope is simpler and at the same time more pragmatic:
given that new theoretical advances or strikingly novel economic policies will not eliminate economic
fluctuations, identifying the variables that influence in a robust way the expected length of a recession can be
of important value for policy making.

Our results point out that more open economies, with deep financial markets, less regulated labor
markets and an overall efficient property rights system experience shorter recessions. We also find important
positive effects coming from GDP growth among trading partners, thereby suggesting that market
diversification is important. At the same time, the evidence suggests a generally better performance of
floating exchange rate regimes as compared to both hard and soft pegs.

The structure of the paper is as follows: section 2 analyses relevant theories and related literature on
the subject. Section 3 presents the methodology used in the paper, both in the description of the data and in
the application of the econometric models. Section 4 presents the most relevant empirical results; the last
section concludes.

2. Theoretical Aspects and Review of the Literature

The economic literature usually analyses business cycles and economic growth as two separate
things, under the assumption that business cycles represent a transitory dynamic force with no effects on long-
term economic growth. Nevertheless, Fatás (2002) has studied the link between business cycles and long-term
growth rates, and reports evidence that cycles cannot be taken always as a temporal deviation from the trend.
He also finds that countries with higher GDP growth volatility tend to grow slower. This effect is far bigger
for developing countries. Similar evidence has been found in Hnatkovska and Loayza (2005). They find that
macroeconomic volatility and long-run economic growth are negatively related, and that this negative link is
exacerbated in countries that are poor, institutionally underveloped, undergoing intermediate stages of

1
See Levine and Renelt (1992), Sala-i-Martin (1997), and Loayza and Soto (2002).
2
See Sachs and Warner (1997) and Sala-i-Martin (2002).

3
financial development, or unable to conduct countercyclical fiscal policies3. Likewise, Aghion et al. (2005)
show that a lower degree of financial development predicts a higher sensivity of growth to exogenous shocks,
as well as a stronger negative effect of volatility on growth.
All this evidence suggests that business cycles and growth volatility is a phenomena not totally independent
of long-term growth.

2.1 What do recovery times depend on?

The question about cross country differences in recovery times from recessions is, in practice, a
question of how countries manage the initial negative shocks that provoke an economic contraction.

Under real business cycle theories4, economic volatility and the allocation of time between expansion
and contraction depends mainly on supply factors; more specifically, on the way in which technological
innovations occur, as they modify the inter-temporal relation of factor remunerations between the present and
the future, as well as the equilibrium between time dedicated to work and leisure. Although this theoretical
line of thought has been criticised5, it has managed to emulate efficiently different economies around the
world. Nevertheless, the procedures behind these methods cannot predict particular events, since they rely on
highly simplified micro-founded models. Additionally, real business cycles models can hardly acknowledge
any differences in recession recovery times across countries because they do not identify the movement of
variables in terms of their mean; instead, they try to assess how well models emulate the real correlations
between the variables of interest.

On the other hand, we can also argue that nominal as well as real market rigidities, jointly with
volatility in aggregate demand, constitute the main factors behind economic volatility. Monetarists support the
idea that monetary shocks, through their incidence on aggregate demand, affect the business cycle and the
time that an economy spends in a recession.

Although the debate between different theories about the business cycle remains, it is possible to
outline some preliminary ideas on why some countries recover faster than others from a recession. There
seem to be three general conditions that explain a large part of the differences between countries’ outcomes:
overall economic openness (trade and financial), the level of factor market rigidities, and the depth of
structural economic reforms have been used to explain recession depth and/or different recovery times.

Certain elements of trade and financial openness can explain differences in the length and severity of
economic downturns. Given the increasing globalization and integration of world markets, international
relationships (both commercial and financial) have an important role in explaining growth volatility. Since
openness is positively correlated with economic growth and international linkages to the world economy are
responsible for smoother consumption and investment paths, more open economies should experience milder
recessions and more stable macro aggregates.

In this line of investigation, Caballero and Krishnamurthy (2002) argue that financial crises are
followed by deeper contractions in investment and growth in those countries with limited access to
international asset markets, i.e. less developed economies. Caballero (2003) addresses this idea more
explicitly through a comparison of the Chilean and Australian experiences after the Asian-Russian crisis of
1997-98. Country selection was due to common characteristics shared by Chile and Australia: both countries
are export-oriented and rely heavily on historically price-volatile commodities. Nevertheless, after the
negative shock in terms of trade due to the crises, Australia recovered much faster than Chile. The Australian
economy did not face a shortage of external capital inflows and the overall adjustment was absorbed by a
higher current account deficit financed by international capital inflows; Chile, in contrast, had a severe

3
See also Loayza et al (2007).
4
Kydland and Prescott (1982), Prescott (1986).
5
See for example, Summers (1986).

4
contraction, both in consumption and output: expenditure fell 15% respect with respect to the pre-crises level
and the economy suffered a recession in 1999. According to Caballero (2003), the overall contraction
experienced by the Chilean economy, was ten times larger than it would have been if the country had stronger
linkages to international capital markets.

The degree of trade openness is the other side of the coin when we talk about the internationalization
of an economy. Deeper trade relations during the second half of the twentieth century are related to the
transmission and characteristics of the international business cycle. Additionally, a diversified export structure
can partially substitute for the lack of access to financial markets, since it provides diversified fund sources
and diminishes the vulnerability to external shocks. However, there is little consensus in the literature about
the empirical relationship between openness and macroeconomic volatility. Buch et. al. (2005) finds no
significant relationship between an increased degree of trade interdependence and macroeconomic volatility.
Likewise, Cecchetti et. al. (2006) find that commercial openness is negatively, but not significantly, correlated
with fluctuations in GDP growth. On the other hand, Karras and Song (1996) and Easterly et. al. (2000) find
that an increase in the degree of trade openness leads to higher output volatility, especially in developing
countries. Finally, others as Barrell and Gottschalk (2004) find evidence that the increase in trade openness
can account for the reduction of the volatility of the output gap in the G7.

During recessions, the economy performs adjustments between sectors, independently of whether the
recession was due to external or idiosyncratic shocks. Trade openness and access to international capital
markets allow the efficient reallocation of consumption and investment inter-temporally. On the other hand,
market flexibility allows the efficient distribution of factors intra-temporally, given a negative shock, thus
minimizing involuntary factor unemployment. For example, analyzing the response of the South Korean
economy to the Asian crises of 1997, Koo and Kiser (2001) found that key factors behind the short recovery
period experienced by that country were the rapid labor market adjustment and the correct set of fiscal
policies used.

Bergoeing et al. (2002) present an example of the first idea when they try to explain why Chile
recovered much faster than Mexico from the debt crisis of the 1980s. Earlier structural reforms in Chile --
they argue-- can account for a large part of the difference in recovery times between both countries. While
Chile attempted deep reforms in trade liberalization, fiscal policy, privatization, financial markets and
bankruptcy legislation throughout the 70s, Mexico only applied this sort of policies late in the 80s. Although
the fall in investment and employment, the debt burden and the exchange rate depreciation were more severe
in Chile, the structural reforms set the basis for faster recoveries and economic growth.

Precise and objective data on the quality of economic structures and institutions is hard to obtain, or
simply does not exist for long periods of time and heterogeneous sets of countries. A similar problem happens
with labor markets. International data sets on labor market rigidities are scarce and do not extend too far into
the past. Given these shortcomings, we focus on trade and openness statistics, which extend both across
countries and through the years.

2.2 A Historical Review of Recessions

Several studies have claimed that the business cycle is getting milder and recessions are getting
shorter for the US economy, among them Zarnowitz (1998), Blanchard and Simon (2001) and Ahmed, Levin
and Wilson (2002). For example, Zarnowitz calculates that the average length of a US recession in the post-
1945 period is 11 months (less than 4 quarters); between 1870 and 1945, on the other hand, that average was
21 months (7 quarters).

It has been argued that this trend extends to the rest of the developed world. According to an IMF
study6 for a sample of 16 rich countries, the average decline of GDP in a recession (between peak and trough

6
IMF World Economic Outlook, May 2002.

5
of the cycle) was 4.3% between 1881-1913; 8.1% between the world wars, and just 2.3% thereafter. Many
hypothesis have been given to account for this fact in the developed world7. Nonetheless, looking at a more
recent time span, Table 1 appears to indicate otherwise: recessions turned longer and more volatile among
OECD countries in the period 1986-95.

Does this pattern apply to less developed countries? This question can be answered only partially,
since long GDP series are usually unavailable for LDCs. A hint is provided by Easterly, Islam and Stiglitz
(2000) who studied a heterogeneous group of countries, both LDCs and OECD members. Their paper
concludes that crises have been getting worse in the last 25 years, especially for LDCs and that the cause of
this lies in volatile financial markets. This apparent asymmetry in the depth of recessions is confirmed by the
following tables, in which information for 51 countries is summarized for the period 1970-2000.

Table 1. Recession statistics, OECD countries

Mean
Number of Standard Country with longest recession
Period life of
recessions deviation (starting year, quarters in recession)
recession
1971-1975 9 3.2 0.97 United States (1974, 5)
1976-1980 6 3.5 1.52 United Kingdom (1980, 6)
1981-1985 14 3.1 1.10 Canada (1982,5) – Portugal (1983,5)
1986-1990 6 5.5 3.62 Finland (1990,12)
1991-1995 9 4.7 2.55 Sweden (1991,11)
Total 44 3.8 2.08

As seen in Table 1, no clear trends relative to the mean life of a recession can be seen for the 1971-
81 sub-sample, though things got worse in 1986-958. The mean life of a recession for the entire sample is 3.8
quarters, very close to the 11 month average calculated by Zarnowitz for the US economy. Note that Table 1
includes the most recession-prone period faced by OECD countries (1981-1985, with 14 recessions) and the
higher dispersion in the length of recessions experienced by developed economies in the late 80s and early
90s.

Table 2. Recession statistics, developing countries

Mean
Number of Standard Country with longest recession
Period life of
recessions deviation (starting year, quarters in recession)
recession
1981-1985 5 6.2 2.59 Philippines (1983,10)
1986-1990 4 4.5 1.29 Peru (1988,6)
1991-1995 13 4.5 3.84 Latvia (1991,13) – Belarus (1993,13)
1996-2000 15 5.3 3.08 Argentina (1998,14)
Total 37 4.95 3.05

Since quarterly GDP series for LDC’s are shorter than those for developed countries, an important
restriction was faced while performing these statistical comparisons. Hence, historical comparisons of GDP
growth volatility and recessions could not be made, as Zarnowitz did for the US economy. Nevertheless, two

7
See Zarnowitz (1998) and Romer (1999).
8
Sample period dependent on data availability.

6
interesting conclusions arise from the tables. First, the average length of a recession is higher for LDCs. Also,
the duration of recessions is more volatile in LDC’s, as noted by the higher standard deviation in Table 2.

It should be noted that the high number of detected recessions in LDC’s in the 90s may be inflated by
the fact that statistics for several transition economies became available only during the 90s; thus, the number
of recessions between 1991 and 2000 may account both for an increasing number of countries with available
statistics and for the occurrence of more recessions per country. Thus, we constructed a third table with Latin
American countries only, for which longer GDP series were available.

Table 3. Recession statistics, Latin American countries

Number of Mean life of Standard Country with longest recession
Period
recessions recession deviation (starting year, quarters in recession)
1981-1985 4 5.3 1.71 Chile (1982, 7)
1986-1990 3 5.0 1.00 Peru (1988, 6)
1991-1995 5 3.0 1.00 Mexico (1994, 4) – Argentina (1995, 4)
1996-2000 5 6.0 4.90 Argentina (1998, 14)
Total 17 4.7 2.75

The Latin American experience lies in between, though much closer to that of developing countries
as a whole: the mean life of a recession is 4.7, between the lower bound set by the OECD countries (3.8) and
the results of the exercise for the whole group of LDCs (4.95). Also, the dispersion of recessions –measured
by the standard deviation of average recession length in the period- lies between the OECD and LDC’s
statistics. Again here, the results for Latin America are much closer to those of LDCs in general than to those
of the OECD economies.

The statistics presented in the tables show a global perspective on the occurrence and trends of
recessions over the last thirty years. Despite the unavailability of longer quarterly GDP series for LDCs,
which would make for richer comparisons, several conclusions can be reached (given our limited sample):
recessions seem to be shorter and less disperse for developed economies; developing countries –especially the
so called transition economies- face longer and more severe recessions.

Thus, although empirical evidence supports the idea of a diminishing trend in the length of the US
economy recessions, this stylized fact cannot be extended to the rest of the world.

3. Methodology
This section presents a general description of the data and methodologies used in the paper. We
identified recessions across countries by studying quarterly GDP series, while the rest of the (explanatory)
variables were used in annual frequency. Our dataset consists of information for 51 countries9, between
1970:I and 2002:IV. A GDP index was used, with the same base year for all countries (1995) in order to make
direct comparisons across series. We defined quarterly GDP growth as the percentage change of the series
over a year ago. This procedure avoids the use of mechanical filters and ad-hoc seasonality adjustments for
individual countries. All series were treated equally, maintaining objectivity and parsimony.

3.1 What is a recession?

The use of the IMF definition of a recession (two or more consecutive quarters of negative GDP
growth) has the virtue of being simple, objective and easy to implement in an international dataset. Of course,

9
The complete list of countries is in the appendix.

7
this is not the only definition of a recession. There are other methods, with distinct requirements and
characteristics.

For example, the recession definition of the National Bureau of Economic Research for the U.S.
economy, takes into account a wide collection of series -namely employment, trade, income and output- and
assesses the fact that there cannot be a complete economic recovery from a recession exclusively when output
is rising: without a correspondent surge in job creation and international trade, the NBER criteria would leave
the U.S. economy still in a recession. Although this approach provides quite a useful amount of information,
its implementation for a wide set of countries lays beyond the scope of this paper. Moreover, specific country
effects could bias the NBER’s procedures; incomplete and unreliable country data are a major hurdle, too.

Another simple way of defining a recession is by estimating output gaps of long-term growth rates,
calculated by means of mechanical filters such as Hodrick-Prescott, Beveridge-Nelson or Baxter-King. Once
produced this estimates for the long-term or “potential” GDP growth series, all negative deviations of the real
data from this trend would represent negative business cycles, or in a capricious sense of the word, recessions.
Since the seminal work by Kydland and Prescott (1982) and the ideas of Zarnowitz (1992), deviations of
filtered series are seen as an accurate depiction of business cycles from peaks to troughs. Nevertheless, this
mechanical algorithm can produce several problems and biases. For example, the distribution of quarters
between recessions and expansions (under and above the trend) can be enormously biased by scalar factors.
Also, mechanical methods have to accommodate to real observations, fictionally creating equal number of
positive and negative deviations. Other frequent shortcomings of applying mechanical filters to time series
data, is that of miscalculation and low consistency of results when the sample length changes and when there
are unitary roots in the series. Additionally, international comparisons would not be feasible nor credible if
the length of the distinct GDP series are gruesomely dissimilar or if there are portions of missing values for
different countries’ datasets, as is the case with our own.

We decided to use the standard definition of a recession mainly because of its simplicity and
objectivity; another important feature is that this is the definition used by the media and authorities in the
majority of countries, thus producing economic policy reactions and movements in local markets.

3.2 Estimation
After identification and calculation of the length of recessions across countries and through time, we
estimated the effect of a group of economic variables over the average length of a recession (in quarters).
Therefore, once the length of a particular recession was attained, the year in which the recession started was
assigned to it. Hence, an annual recession vector was constructed10 which was attached to the annual
frequency matrix of explanatory variables.

Given the way in which the dependent variable is constructed, it represents a strictly positive integer
number that fits the characteristics of count-data models, used mainly in microeconomic analysis. The number
of quarters an economy spends in a recession hardly resembles a “normal” distribution, as seen in figure 1.

Figure 1. Histogram and Statistics of Recession

10
The annual series contain zeros (years without recessions), positive integer numbers (years with
recessions), and non-available observations (years following a recession, since a recession can span
consecutive years).

8
 30
Observations 77
25
Mean 4.8
Median 4.0
20 Maximum 14.0
Minimum 3.0
Std. Dev. 2.5
15
Jarque-Bera 133.73
10 Probability 0.0000

0
4 6 8 10 12 14

length of recession in quarters

It is evidently from the graph that the large part of the observed outcomes present themselves in low
frequencies. The majority of recessions (in our sample) lasts between 2 and 4 quarters –mean life of a
recession is 4.8 quarters- and very few have durations exceeding 14 quarters. On the other hand, the Jarque-
Bera normality test rejects the null hypothesis that the length of recessions follows a normal distribution at a
confidence level of 99%.
According to this evidence, we applied count-data models to perform the empirical analysis11.
Specifically, we use the Poisson distribution: this is the standard distribution in this type of regressions and
the empirical distribution follows closely the theoretical one, as seen in figures 2 and 3.

Figure 2. Theoretical Poisson Distribution

.35
P ( x = X ) = m x e − m x!
.30
m=1.5 m=2.5 m=3.5
.25

.20

.15

.10

.05

.00
2 4 6 8 10 12 14

Figure 3. Length of Recessions’ Empirical Distribution

11
Specifics of this type of models can be seen in the appendix.

9
 .35

.30

.25

.20

.15

.10

.05

.00
25 50 75 100

length of recessions: empirical density

Figure 3 depicts the kernel density estimate of the distribution of the recession series, using an
Epanechnikov weight and a hundred grid points. As said before, it follows closely the theoretical distribution,
like the one in Figure 2, created with 15 observations and varying mean parameters. The comparison of both
figures is straightforward: they show that a Poisson distribution fits well the data and that the estimation of a
count-data model for this macro-phenomena is not far-fetched.

In order to assure weak exogeneity of regressors, special attention was put in introducing non-
contemporary variables at the RHS of the equation. Endogeneity of explanatory variables invalidates the weak
exogeneity assumption and produces non-robust and inconsistent estimators. Hence, most regressors were
introduced in lagged terms: variables explaining the length of a recession which started at year t, are from
year t-1, unless strong a-priori belief exists in order to treat them as exogenous and thus, introduce them
contemporarily into the regression.

3.3 Explanatory Variables

This section presents a detailed discussion of the independent variables used in our study. All
variables are in lagged form, except the contrary is explicitly stated. As explained before, the variables are
focused in measuring overall economic openness due to availability and quality of international data.

• Trading partners’ GDP per capita growth. Represents the annual percentage change of trading partners’
GDP per capita growth, weighted by trade share. This variable was treated as exogenous (contemporary).
Its expected sign is negative: countries with more dynamic trading partners should experience milder
recessions.

• Real exchange rate misalignment. The absolute value of the difference between the actual real exchange
rate and its long-term trend, given by a Hodrick-Prescott filter. The relationship between RER
misalignment and length of recessions is supposed to be positive, because larger misalignments are
associated to longer adjustments.

• Terms of trade shock. The annual percentage change of the terms of trade series, the year before the start
of a recession. Since worsening terms of trade represents a situation of economic weakness, the expected
sign of the coefficient is negative: growing (decreasing) terms of trade determine shorter (longer)
recessions.

• Trade openness. As a standard procedure, this variable is measured as the sum of exports and imports, as
a fraction of GDP. This particular definition of trade openness is superior to other alternatives in the

10
 sense that it captures efficiently the existence of non-tariff barriers to foreign trade. We expect a negative
sign for this coefficient, since more open countries are able to smooth out negative shocks easily, thus
experience milder recessions.

• Exchange rate regimes. We include dummy variables for the case of floating exchange rates and hard
pegs. The effect of soft pegs (intermediate exchange rates) is captured by the constant of each regression.
We use the de-facto classification by Levy-Yeyati and Sturzenegger (2002), since this classification stays
true to what countries actually do rather than on what countries say they do (“deeds versus words” ).
Although the expected sign of the coefficient is ambiguous, some international evidence tends to show
that flexible exchange rates are related to better macroeconomic outcomes and more flexibility amidst
negative shocks, in relation to soft/hard pegs .

• GDP per capita growth the year the recession starts. We include this variable in order to control for the
overall severity of the economic downturn. We expect this coefficient to be negative, since a greater
annual growth rate during a recession year is linked with shorter recessions.

• Domestic credit provided by the banking sector (as percentage of GDP). Following Caballero (2003) and
Caballero and Krishnamurthy (2002), we control for the effect of different levels of financial
intermediation in the economy. Countries with stronger financial markets (higher domestic credit), should
be prone to shorter recessions due to their ability to smooth out real shocks through future periods; hence,
this coefficient is expected to be negative.

• Persistence of Unemployment. This variable is constructed as the simple correlation coefficient between
unemployment in period t vs. period t-1 for country i, using a rolling window sample of five years.
Despite its simplicity, this variable attempts to capture the extent of regulatory burden in the labor
markets, which influences periods of high and persistent unemployment. Countries with less flexible
labor markets should experience longer recessions because of the difficulties to resume production after a
negative real shock; thus, we expect a positive coefficient for this variable.

• Property rights index. Based on disaggregate information of the Global Competitiveness Report, the
International Country Risk Guide and the Economic Freedom of the World Index. Our measurement of
the quality of property rights and the overall judiciary system in each country spans from 1970 to 2000,
and tries to assess the importance of an efficient legal system at times of economic depression, during
which the creation and destruction of private businesses is more important. Better legal frameworks, with
efficient procedures to restructure firms is expected to be negatively related to the length of a recession.

4. Results

4.1 Poisson regression results

This section presents the most relevant empirical results. According to the discussion in section 3.2,
we estimate a Poisson regression, in which the dependent variable is the length of a recession (in quarters) and
the explanatory variables are the annual observations of the variables discussed in the previous section.

Using standard notation for dependent (y) and explanatory (x) variables, Poisson-type regressions are
based in the conditional density of y given x12:

12
Note that temporal indexes have been eliminated, in order to facilitate exposition and because in the
different count-data models estimated throughout the paper, observations where pooled.

11
 e − m ( xi , β ) m( xi , β ) yi
f ( y i | xi , β ) = (1)
yi !

Usually, a prior assumption regarding the conditional mean parameter (m) must be made. The standard
formulation (used in this paper) is the log-linear:

m( xi , β ) = E ( yi | xi , β ) = exp( x'i β ) (2)

The vector of coefficients β is calculated by maximizing the log-likelihood function

N
l( β ) = ∑ y i Ln[m( x i , β )] − m( xi , β ) − Ln( y i !) (3)
i =1

by means of an iterative numeric method (Newton-Raphson, Bernd-Hall-Hall-Hausman), given the non-

linearity of the first order conditions set by objective function.

Our results are in line with the previous analysis of explanatory variables, in terms of expected signs
of coefficients and overall statistical significance. They support the idea that more open countries, with better
legal frameworks and more flexible labor markets suffer from shorter recessions, i.e., recover faster from
negative output shocks. The benchmark regression is shown in Table 4.

As seen in the table, all variables present the expected sign and the overall regression shows high
significance. According to the LR statistic, the null hypothesis of global insignificance is rejected with 99%
confidence. Table 4 shows that all but the fixed exchange rate regime variable are statistically significant.

These results have strong implications in light of the literature on business cycles and recoveries.
The main general finding is that overall rigidities in the economy tend to be inefficient at times of
recessions. Deep financial systems, low levels of labor market rigidities (proxied by our unemployment
persistence index) and the quality of property rights in a given country favor faster recoveries. These
variables relate to a more general idea, which is the ability of an economic system to smooth out shocks
through time and reallocate resources across different sectors. This rationalization is quite direct: deeper
financial systems, which provide higher amounts of domestic credit to the private sector allow for output
shocks to be more easily smoothed out during economic fluctuations (when no liquidity shortages are
present); labor market rigidities tend to produce inefficient equilibria in domestic markets, slowing
recoveries; finally, the overall state of property rights is an important determinant of economic recovery
when a process of resource reallocation is taking place; high explicit and/or implicit costs of discarding
inefficient firms at the trough of the business cycle may have important aggregate effects in terms of the
length of a recession.

Table 4. Poisson Regression

Dependent variable: number
of quarters in recession
GDP pc growth of trading partners -0.421239***
(-13.2822)

Real exchange rate misalignment 1.076577**

(2.01258)

Terms of trade shock -4.422191***

(-4.703467)

12
 Trade Openness -0.396844***
(-2.619378)

Exchange rate regime (d=1 if fixed) -0.044103

(-0.402388)

Exchange rate regime (d=1 if float) -0.253788***

(-2.59574)

GDP per capita growth (current year) -0.097911***

(-18.43701)

Unemployment Persistence 0.513282***

(5.069766)

Domestic Credit -0.003829***

(-2.812978)

Property Rights Index -0.08938***

(-2.993934)
Considered recessions (Obs.) 52 (315)
LR index (pseudo R2) 0.202
LR statistic p-value† 0.0000
Notes: Intercept not shown; z-statistics in parenthesis.
† Null hypothesis: all coefficients equal zero.
*, ** and *** represent significance at 10, 5 and 1% respectively.

According to our results, trade related variables are also important determinants of the length of a
recession. On the one hand, GDP growth among trading partners has a negative, statistically significant
coefficient, which reinforces the value of market diversification for a country’s exports. Also, trade openness
(the sum of imports and exports as a percentage of GDP) has an analogous effect on recession length, which
suggests a trade-manner of defusing negative output shocks: in simple, trade in goods de-couples trends in
output from trends in consumption and investment. On the other hand, real exchange rate misalignments and
negative terms of trade shocks produce longer recessions, a result hinting the importance of maintaining
realistic prices and external equilibrium during economic slowdowns.

As discussed before, the expected sign of the exchange rate regime coefficients is ambiguous ex-
ante. There is some empirical evidence on the macroeconomic performance of different regimes, but there is
no theoretical consensus supporting any particular exchange rate regime above the rest. The results presented
in Table 4 are in line with some of the previous literature relating the choice of exchange rate regimes with
macroeconomic volatility13. Both dummy variables (floating regimes and hard-pegs) show negative
coefficients; this results indicate that soft-pegs (or intermediate regimes) are the worse performers in our
sample: those countries managing soft-pegs are more likely to spend more quarters in recession. Our results
tend to favour floating exchange regimes, since its coefficient is statistically significant (as opposed to the
hard-peg dummy). Thus, according to the evidence, countries with floating exchange rate regimes would tend
to have shorter recessions.

4.2 Robustness

13
See, for example, Larrain and Parro (2003), and Larraín (2005).

13
 In order to test the robustness of our previous results, we estimate the same equation under two
alternative methods: a count-data regression model, under a different distribution assumption (Negative-
Binomial distribution) and a seemingly unrelated regressions (SUR) approach14.

Negative- Binomial and SUR

The Negative-Binomial distribution case is of interest here, since it allows for over-dispersion in the
dependent variable (as is usually the empirical case), something that can produce inconsistent estimators in
Poisson regressions. The Negative-Binomial distribution can be seen as a more general specification, from
which the Poisson distribution is a particular case. Specifically, the log-likelihood function changes to:
n
l( β , η ) = ∑ y i Ln(η 2 m( x i , β )) − ( y i + 1 )Ln(1 + η 2 m( x i , β )
i =1
η2
(4)
+ LnΓ( y i + 1 2 ) − Ln( y i ! ) − Ln Γ(
1 )
η η2
2
where η it’s the over-dispersion parameter, estimated jointly with the coefficient vector β. As discussed
above, the function is maximized through an iterative method, while the definition of conditional mean (m)
remains.

Seemingly unrelated regressions were also tested because this method estimates the parameters of a
system of equations, accounting for heteroskedasticity and contemporaneous correlation in the disturbances
across equations. Results from both alternative regressions provide us with a reasonable robustness test, given
the dissimilarity of estimation procedures/techniques. Close coefficient estimates along similar goodness of fit
across regressions would indicate robustness of our results under different estimation procedures. Table 5
presents these results.

Table 5. Alternative Estimation Methods

Dependent variable: number of
quarters in recession
Neg-Bin SUR
GDP pc growth of trading partners -0.499549*** -0.391619***
(-11.05201) (-21.26463)

Real exchange rate misalignment 1.053786 0.368823

(1.505012) (1.597975)

Terms of trade shock -4.562891*** -3.30046***

(-3.451855) (-7.155357)

Trade Openness -0.401413* -0.002551***

(-1.872595) (-2.753689)

Exchange rate regime (d=1 if fixed) -0.11271 -0.102217*

(-0.720475) (-1.924717)

Exchange rate regime (d=1 if float) -0.2275* -0.15539***

(-1.736019) (-3.959241)

14
See Zellner (1962).

14
 GDP per capita growth (current year) -0.098968*** -0.045599***
(-15.89476) (-8.449199)

Unemployment Persistence 0.482566*** 0.18213***

(3.53278) (3.987892)

Domestic Credit -0.004309** -0.001805***

(-2.253183) (-3.590748)

Property Rights Index -0.10229** -0.074709***

(-2.450388) (-3.754116)
Considered recessions (Obs.) 51 (315) 51 (315)
Goodness of Fit‡ 0.544 0.14
LR statistic p-value† 0.0000 --
Notes: Intercepts not shown. Standard errors in parenthesis. Significance estimated through z-statistics in
the Negative-Binomial case and t-statistics in the SUR case.
* For Negative-Binomial: LR index; for SUR: Adjusted R2
† Null hypothesis: all coefficients equal zero.
*, ** and *** represent significance at 10, 5 and 1% respectively.

As seen in the table, the robustness of our earlier estimates is confirmed, taking into account the
goodness-of-fit and the overall significance statistics. Notice that in the negative-binomial regression, the LR
index is significantly higher than in all other cases (0.54 vs. 0.14 and 0.20 of the SUR and Poisson
approaches). This better adjustment follows from the way the model is estimated along the extra coefficient,
which accounts for over-dispersion in the dependent variable.

In terms of individual coefficients, the robustness exercise presented above shows mixed results;
most variables maintain high statistical significance and their expected signs. On the other hand, the real
exchange rate misalignment looses significance under both alternative regressions, while the openness
variable underscores its previously high significance. Curiously, it seems that only the trade related variables
suffered from important alterations between estimation methods.

Under both regressions, the rest of the explanatory variables remained highly significant. This is the
case of growth among trading partners, unemployment persistence, domestic credit and property rights.
These confirms our prior beliefs discussed in the previous section. Despite the dissimilarity of these
estimation methods, there remains a strong case in favor of the flexibility hypothesis, i.e., those countries with
less rigidities (ability to reallocate resources through the financial system and across sectors) experience
shorter recessions.

Results for the exchange rate regimes support the conclusions remain from the last section.
Intermediate regimes (or soft-pegs) present the worst performance among the three regime options, in terms
of additional quarters in recession. Furthermore, the evidence ratifies the superiority of flexible exchange
rates versus the other two regimes. This result is maintained both in the Negative-Binomial regression and in
the SUR approach, although in the latter regression the fixed exchange rate dummy becomes marginally
significant.

It must be noted that the coefficients estimated by count-data models and those estimated under SUR
are not directly comparable, in terms of the effect on the dependent variable, because of the non-linearity of
count-data models. Nevertheless, direct comparisons can be made through “mean marginal effects”, as
discussed below.

15
 Given the expression used to estimate the dependent variable conditional mean, an expression for
each observation can be found:

m( xi , βˆ ) = exp( x' i βˆ ) = mˆ i (5)

then
∂m( x i , βˆ ) ˆ
Marginal Effectk = ∂β k = exp( x' i β )β k = m
ˆ iβk (6)

In other words, the effect of each RHS variable over the conditional mean of y is sample variant, i.e.
can be calculated for each observation “i”. In order to simplify the discussion and the presentation of results,
mean marginal effects are shown in Table 6, and calculated as follows:

Mean Marginal Effectk = exp( x ' i βˆ )β k = constant (7)

The table enables a direct comparison of the calculated effects of the different explanatory variables
on the length of a recession, measured in quarters. Confirming the earlier discussion, results from the table
show significant marginal effects of GDP per capita growth of trading partners, trade openness, exchange rate
regime choice, property rights, unemployment persistence and domestic credit. Across methods, the two
count-data models are closely related (Poisson and Neg-Bin columns) in terms of calculated marginal effects,
while the SUR methodology seems to diverge somewhat. However, the differences under all methods are
quite minimal for specific coefficients, such as in the growth of GDP per capita of trading partners, the
openness variable, the persistence of unemployment and the domestic credit variables.

Table 6. Mean Marginal Effects Comparison

∂F ∂x i

Poisson Neg-Bin SUR

GDP pc growth of trading partners -0.1914 -0.2110 -0.3916

Real exchange rate misalignment 0.4892 0.4450 0.3688

Terms of trade shock -2.0094 -1.9270 -3.3005

Trade Openness -0.1803 -0.1695 -0. 2551

Exchange rate regime (d=1 if fixed) -0.0200 -0.0476 -0.1022

Exchange rate regime (d=1 if float) -0.1153 -0.0961 -0.1554

GDP per capita growth (current year) -0.0445 -0.0418 -0.0456

Unemployment Persistence 0.2332 0.2038 0.1821

Domestic Credit -0.0017 -0.0018 -0.0018

Property Rights Index -0.0406 -0.0432 -0.0747

16
 Significant effects at 10% or more in italics.

5. Concluding Remarks

This paper has tried to answer the question of why some countries take longer than others to get out
from a recession. Our results from a count-data model under a Poisson distribution show that open
economies, with floating exchange rates, deep financial markets, less rigid labor markets and an efficient
property rights system spend less time in recession. These conclusions remain under different specifications
and procedures, a sign of robustness of our estimates.

Specifically, we tested the role of the following variables in determining the expected length of a
recession –measured in quarters, according to the IMF definition: per capita GDP growth of a country’s
trading partners, real exchange rate misalignment, terms of trade shocks, trade openness, exchange rate
regime, persistence of unemployment, domestic credit provided by the private banking sector, a property
rights index and the GDP per capita growth rate of the country the same year the recession starts (to control
for the magnitude of the downturn). In general, coefficients show the expected signs and are statistically
significant. Thus, countries with the ability to reshape their economic structures more rapidly are able to
recover faster from a recession. These countries share similar characteristics: diversified markets for their
exports, high levels of trade openness, low-persistence unemployment, deep financial market and an efficient
judiciary system, able to enforce well-defined property rights.

The choice of the exchange rate regime is also important. Our econometric results imply that
countries managing a flexible exchange rate regime at the start of a recession, spend less quarters in recession
than those with any other regime choice. The worst performing regimes are the soft-pegs or intermediate
regimes.

17
Appendices
Appendix 1: List of Countries

ARGENTINA, ARMENIA, AUSTRALIA, AUSTRIA, BELARUS, BELGIUM, BOTSWANA,

BRAZIL, BULGARIA, CANADA, CHILE, COLOMBIA, CZECH REPUBLIC, DENMARK,
ECUADOR, ESTONIA, FINLAND, FRANCE, GERMANY, GREECE, HONG KONG, HUNGARY,
IRAN, IRELAND, ISRAEL, ITALY, JAPAN, KAZAKHSTAN, (SOUTH) KOREA, LATVIA,
LITHUANIA, MALAYSIA, MALTA, MEXICO, NAMIBIA, NETHERLANDS, NEW ZEALAND,
NORWAY, PERU, PHILIPPINES, POLAND, PORTUGAL, SLOVAK REPUBLIC, SPAIN, SWEDEN,
SWITZERLAND, THAILAND, TURKEY, UNITED KINGDOM, UNITED STATES, VENEZUELA.

Appendix 2: Data sources

Variable Source
Quarterly GDP index, 1995 =100. International Financial Statistics (IFS) CD-ROM
International Monetary Fund (IMF)
Trading Partners' GDP Per Capita growth IMF: Directions of Trade
(%, weighted average by trade share) Global Development Finance
World Development Indicators

Real Effective Exchange Rate Index Global Development Finance

World Development Indicators

Terms of Trade index (goods and Global Development Finance

services), 1995=100 World Development Indicators

Trade as percentage of GDP (sum of Global Development Finance

imports and exports of goods and World Development Indicators
services as fraction of total output)

GDP per capita growth Global Development Finance

World Development Indicators

Unemployment (% of total Labor Force) International Financial Statistics (IFS) CD-ROM

International Monetary Fund (IMF)

Domestic Credit Provided by Banking Global Development Finance

Sector World Development Indicators

Property Rights Index Economic Freedom of the World (2003) http://www.fdsa??.fd

Based on information of the Global Competitiviness Report and the
International Country Risk Guide.

De-facto classification of exchange rate Eduardo Levy-Yeyati’s homepage, Universidad Torcuato di Tella:
regimes http://www.utdt.edu/~ely/base_2002.xls

18
References
Aghion, P., Angeletos, M., Banerjee A., and K. Manova. “Volatility and Growth: Credit Constraints and
Productivity-Enhancing Investment.” NBER Working Paper No. 11349. 2005.

Ahmed, S., Levin A. and B.A. Wilson. “Recent U.S. Macroeconomic Stability: Good Policies, Good
Practices or Good Luck?” Federal Reserve Board IFD paper No. 730. 2002.

Barrell R and S Gottschalk. “The volatility of the output gap in the G7.” National Institute Economic
Review 188, 2004, pp. 100–107.

Barro, R. and X. Sala-i-Martin. “Convergence.” Journal of Political Economy 100(2), 1992, pp. 223-251.

Bergoeing, R., Kehoe, P., Kehoe, T. and R. Soto. “Policy-Driven Productivity in Chile and Mexico in the
1980s and 1990s.” American Economic Review 92(2), 2002, pp. 16-21.

Blanchard, O. and J. Simon. “The Long and Large Decline in U.S. Output Volatility.” Brooking Papers
on Economic Activity 1. 2001, pp. 135-174.

Buch, C.M., J. Döpke and C. Pierdzioch. “Financial openness and business cycle volatility.” Journal of
International Money and Finance 24(5), 2005, pp. 744-765.

Caballero, R. “Coping with Chile’s External Vulnerability: A Financial Problem.” In Loayza, N. and R.
Soto (eds.). Central Banking, Analysis and Economic Policies, Vol. 6, Central Bank of Chile:
Santiago, 2003, pp. 377-415.

Caballero, R. and A. Krishnamurthy. “A Dual Liquidity Model for Emerging Markets.”, American
Economic Review, Papers and Proceedings 92(2), 2002, pp.33-37.

Cecchetti, S., Flores-Lagunes A., and S. Krause. “Assessing the Sources of Changes in the Volatility of
Real Output.” The Economics Journal 116(4), 2006, pp. 408-433.

Easterly, W., Islam, R. and J.E. Stiglitz. “Shaken and Stirred: Explaining Growth Volatility.” Annual
World Bank Series of Development Economics. World Bank: Washington DC, 2000.

Fatás, A. “The Effects of Business Cycles on Growth.” In Loayza, N., Soto, R. (eds.) Central Banking,
Analysis and Economic Policies, Vol. 6, Central Bank of Chile: Santiago, 2002, pp.191-219.

Hnatkovska, V. and N. Loayza. “Volatility and Growth.” In Aizenman J., Pinto B. (eds.) Managing
Volatility and Crises: A Practicioner’s Guide. Cambridge: Cambridge University Press, 2005.

IMF. “Recessions and Recoveries.” In IMF World Economic Outlook, Chapter 3, IMF: Washington DC,
2002, pp.104-137.

Karras, G., and F. Song. “Sources of business-cycle volatility: an exploratory study on a sample of
OECD countries.” Journal of Macroeconomics 18, 1996, pp.621–637.

Kydland, F. and E. Prescott. “Time to Build and Aggregate Fluctuations.” Econometrica 50(6), 1982, pp.
1345-1370.

Koo, J. and S. Kiser. “Recovery From a Financial Crisis: The Case of South Korea.” Economic &
Financial Review Q IV, 2001, pp. 24-36.

19
Larraín, F. “Flotar o Dolarizar: ¿Qué Nos Dice la Evidencia?” El Trimestre Económico, vol. LXXII (1),
january-march 2005, pp. 5-28.

Larraín, F. and S. Choi. “Recesiones, Apertura y Régimen Cambiario.” Cuadernos de Economía, Latin
American Journal of Economics 40(121), 2003, pp. 668-678.

Larraín, F. and F. Parro. (2003). “¿Importa la Opción de Régimen Cambiario? Evidencia Para Países en
Desarrollo”. Cuadernos de Economía, Latin American Journal of Economics.

Larraín, F. and A. Velasco. (2001). “Exchange-rate Policy in Emerging Market Economies: The Case for
Floating.” Essays in International Economics N°224. Princeton University: Princeton, NJ.

Levine, R. and D. Renelt. (1992). “A Sensitivity Analysis of Cross-Country Growth Regressions.”

American Economic Review 82(4), 942-963.

Levy-Yeyati, E. and F. Sturzenegger. (2002). “Classifying Exchange Rate Regimes: Deeds vs. Words.”
Mimeo, Torcuato Di Tella University.

Levy-Yeyati, E. and F. Sturzenegger. (2001). “Exchange Rate Regimes and Economic Performance.”
IMF Staff Papers 47.

Loayza, N. and R. Soto. (2002). “The Sources of Economic Growth: An Overview” In Loayza, N., Soto,
R. (eds.), Central Banking, Analysis and Economic Policies, Vol. 6 Central Bank of Chile:
Santiago, 1-39.

Loayza, N., Ranciere, R., Serven, L. and J. Ventura (2007). “Macroeconomic volatility and welfare in
developing countries: an introduction”. World Bank Economic Review 21, pp. 343-357, April.

Prescott, E. (1986). “A Theory Ahead of Business Cycle Measurement.” Federal Reserve Bank of
Minneapolis Quarterly Review, Autumn.

Romer, C.D. (1999). “Changes in Business Cycles: Evidence and Explanations.” Journal of Economic
Perspectives, American Economic Association, vol. 13(20).

Sachs, J. and A. Warner. (1997). “Fundamental Sources of Long-Run Growth.” American Economic
Review, Papers and Proceedings 87(2), 184-188.

Sala-i-Martin, X. (2002). “Fifteen Years of New Growth Economics: What Have We Learnt?” In
Loayza, N., Soto, R. (eds.), Central Banking, Analysis and Economic Policies, Vol. 6 Central Bank
of Chile: Santiago, 41-59.

Sala-i-Martin, X. (1997). “I Just Ran Two Million Regressions” American Economic Review, Papers and
Proceedings 87(2), 178-183.

Stock, J. and M. Watson. (2002). “Has the Business Cycle Changed and Why?” NBER 17th Annual
Conference on Macroeconomics.

Summers, L.H. (1986). “Some Skeptical Observations on Real Business Cycle Theory.” Federal Reserve
Bank of Minneapolis Quarterly Review, Autumn.

Zarnowitz, V. (1992). Business Cycles: Theory, History Indicators and Forecasting. University of
Chicago Press: Chicago, Ill.

Zarnowitz, V. (1998). “Has the Business Cycle Been Abolished?” Business Economics 33(4), 39-45.

20
Zellner, A. (1962). “An Efficient Method of Estimating Seemingly Unrelated Regressions and Tests for
Aggregation Bias.” Journal of the American Statistical Association 57, 348-368.

21